Repository navigation
frontproxy indexing of deleted logicalclusters has a large delay #4073
Description
When deleting logical clusters, there is a multi-minute delay before it is picked up by front-proxy. This leads to kubectl get workspace requests to still display workspaces as ready, even though you cannot interact with them anymore. For example in the bottom right, you can see it takes ~9 minutes before indexing even triggers and afterwards it is lacking behind by minutes still.
Activity
mjudeikis commentedon Apr 27, 2026
Did you checked this is still a case in main? I think I fixed it here #4047
SimonTheLeg commentedon May 12, 2026
yes this still occurs, here is a screenshot running
gf79ae83a046a6efrom main branchxrstf commentedon Jun 5, 2026
I am trying to reproduce this, but am failing both on the current
mainbranch as well as onrelease-0.30. I tried to create 50 workspaces and deleting them all, which worked without issues. Then i created 50 workspaces on the root shard, picked a random one that was scheduled onto shard-1, then created 50 more workspaces in there and then deleted all workspaces recursively (from the root workspace again). Again, metrics were updated immediately.SimonTheLeg commentedon Jun 8, 2026
I have re-evaluated this. While I do think it is a bit strange that the indexed and actual logicalclusters red and orange lines seem to close in every 5 minutes, the initial delay is completely gone now and this behavior also only happens in the beginning. Towards the end both are similar. Closing this issue now.